光伏池塘环境中 ,传统无人船定位依赖卫星系统易受到光伏板的遮挡 ,定位精度出现明显下降 ;而基于点特征的视觉同步定位与建图技术也存在轨迹漂移和误差累计等问题 。本研究提出了一种基于光伏池塘环境的融合点线结构特征的双目视觉惯性 SLAM 系统 ,对光伏池塘中线特征和结构特征的提取进行研究 。为 了应对水面反射和光照变化 ,提高线特征质量 ,本研究改进了 ELSED 线特征提取方法 。为了适应光伏池塘环境 ,设计了椭圆球面栅格聚类消隐点 ,并在椭圆球面上使用RANSAC 提取结构特征 。使用开源USVInland 数据集和私有光伏池塘数据集试验验证 ,将本研究 SLAM 系统与改进前 ORB-SLAM3 对比 ,试验结果显示轨迹误差明显下降 ,降幅达到 60% , 系统跟踪帧率达到34 帧/s,保证了系统实时性 ,定位精度满足光伏池塘无人船需求。

In the photovoltaic pond environment,the positioning of traditional unmanned ships relies on satellite systems,which are easily blocked by photovoltaic panels, resulting in a significant decline in positioning accuracy. However, the visual synchronous positioning and mapping technology based on point features also has problems such as trajectory drift and error accumulation. Therefore,this paper proposes a binocular vision inertial SLAM system that integrates point - line structural features based on the photovoltaic pond environment,and studies the extraction of centerline features and structural features of the photovoltaic pond. To cope with water surface reflection and illumination changes and improve the quality of line features, this paper improves the ELSED line feature extraction method. To adapt to the environment of photovoltaic ponds,this paper designs elliptic-spherical raster clustering for vanishing points and uses RANSAC to extract structural features on the elliptic- spherical surface. Experimental verification was conducted using the open - source USVInland dataset and the private photovoltaic pond dataset. By comparing the SLAM system in this paper with the ORB - SLAM3 before improvement,the experimental trajectory error decreased significantly by 60% ,and the system tracking frame rate reached 34 frames per second, ensuring the real - time performance of the system. The positioning accuracy met the requirements of unmanned boats in photovoltaic ponds.